Investigation and identification of functional post-translational modification sites associated with drug binding and protein-protein interactions

Mapping of PTM sites to the tertiary structures of proteins

To identify the spatial composition of PTM substrate sites within the tertiary structures of proteins, we obtained from the PDB protein structures that have been determined by NMR or X-ray crystallography with an experimental resolution less than 2.5 Å [24]. According to the annotations in UniProtKB, 23,605 proteins in the PDB have 3D structure information. Also, chemical groups that can be covalently attached to the side chain of target residues were observed only in a few protein structures. Thus, to locate PTM substrate sites in 3D, mapping was performed between experimentally verified PTM peptides and the PDB protein records, and cross-referenced with the annotations of tertiary structures on UniProtKB with 100% similarity in sequence identity. Additionally, PTM sites possessing protein structures with modified residues were obtained from annotations on the PTM-SD database [19]. Most PTM sites that are mapped to structural sites are presented in the unmodified state, but PTM-SD provides complete information for modified PTM sites in 3D structures.

Investigation of PTM sites associated with drugs binding

While it is suggested that the binding affinity of a small molecule can be regulated by a phosphorylation site within 12 Å of the site of binding [24], there is still a lack of information regarding the occurrences and influence of PTMs near drug-target binding. Therefore, we proposed a method in this study to identify PTM sites involved in drug binding. Figure 1 illustrates the workflow for extracting sites of drug-target binding in protein 3D structure. The entire process can be divided into two steps: 1) the processing of experimentally verified binding sites, and 2) molecular docking of drug binding. In step 1, we collected the structural information of small molecules that have associated keywords such as “drug,” “inhibitor,” “agonist” or “antagonist” and have drug annotations in the DrugBank [35]. A total of 34,555 PDB structures and 4803 small drug molecules which have DrugBank annotations were obtained. Then, the PoseView [36] method was employed to check the binding sites of each drug in the target proteins. PoseView provides a two-dimensional (2D) diagram showing how the drug ligand and the amino acid residues of the target protein may be arranged at the site of interaction. The nature of the interaction is presented in three ways. Black dashed lines indicate hydrogen bonds, salt bridges, and metal interactions. Green solid lines show hydrophobic interactions and green dashed lines represent π-π and π-cation interactions.

In step 2, a docking program, iGEMDOCK 2.0 [37], was utilized for the computational extraction of drug binding sites. We followed the four sequential steps in iGEMDOCK to perform the drug-target interaction analysis: target and database preparations, molecular docking and post-docking analyses. First, coordinates of the target protein atoms acquired from PDB, the ligand binding area, the atom’s formal charge and the atom types were specified. This procedure allowed iGEMDOCK to read the atom coordinates of a ligand from the prepared ligand database. After the ligand database and the target proteins were determined, docking was analyzed for each ligand using the flexible docking function provided by iGEMDOCK. The final step constituted the re-ranking and sorting of all docked ligand conformations based on an empirical scoring function and an evolutionary approach. The output of the program consisted of details regarding the docking result of each binding site, as well as the atomic characteristics of the target residues that interact with a specific drug ligand by hydrogen bonding (H), electrostatic (E) and van der Waal contact (V). A total of 1991 approved drugs from the DrugBank with 1632 target proteins were investigated by this proposed method. After mapping the experimentally verified PTM sites to the PDB structures, the PTM sites located in a drug binding site were determined to have strong associations with drug-binding, while those in the side chains that were within 12 Å of a drug-binding site were considered to be have relatively weak association with drug binding.

Identification of PTM sites related to protein-protein interactions

In this work, the information of protein functional domains and PPIs were integrated for the identification of PTM-dependent protein interactions. To investigate the preferred functional domains of PTMs, we extracted the domain annotations from the Pfam database, which gives protein “signatures” based on protein families, domains and functional sites. In order to comprehensively study the structural properties of PTM sites associated with protein-interaction domains, the 3D structures of PPIs were acquired from the PDB. By adopting the 3D Interacting Domains (3DID) method proposed by Mosca et al. [33], the interaction interface of domain-domain interactions in the PDB 3D structures were determined as illustrated in Fig. 1. First we searched for protein structures with more than two subunits, and calculated the number of contact residues on the interface of the Pfam domain region containing the two subunits. Next, we applied a method based on previously published literature by Aloy and Russell [38], in which they derived the main-chain to side-chain and side-chain to side-chain potentials from the type of complexes described above. In particular, Aloy and Russell [38] defined interacting residues by using one or more of the following properties: hydrogen bonds (N-O distance of 3.5 Å), salt bridges (N-O distance of 5.5 Å), or van de Waals interactions (C-C distance of 5 Å). If there exists more than five pairs of residue contacts between two domains of a two-subunit region, these two subunits would be defined as an interaction structure. The contact residues were also extracted. A total of 30,455 PDB structures and 13,645 proteins were analyzed and 15,124 protein-protein interaction pairs were defined. Mapping between the experimentally verified PTM sites and the PDB structure uncovered PTM sites located on the PPI interfaces. These sites were regarded as PTM-driven PPIs.

PTM substrate site characterization

The availability of high-throughput proteomic technology has stimulated interests in understanding the structural environment of PTM substrate sites [17, 39]. To characterize PTM substrate sites, we considered a five-step approach, focusing on protein properties such as spatial amino acid composition, structurally neighboring residues and side chain orientations surrounding the PTM substrate sites, as well as the secondary structure, division of acidity and alkaline residues, and solvent-accessible surface area. In particular, we adopted the dictionary of protein secondary structure (DSSP) [40] for the calculation of solvent-accessible surface areas of proteins and for the standardization of PDB secondary structures with the corresponding PTM sites. To overcome difficulties that may arise during the derivation of substrate motifs from linear sequences [41], a radial cumulative propensity plot [42] was used to display the spatial composition and abundance of amino acids within and surrounding a particular PTM site (Fig. 2a). After a comprehensive and systematic analysis on the PDB structures, the number of PTM sites that can be mapped on protein structures is presented in (Additional file 2: Table S2). The spatial amino acid compositions were obtained by computing the relative frequencies of the 20 amino acids within 2 to 12 Å radial distances of the modified residues. Next, using JSmol software [43], neighboring amino acids at the sequence level and in the spatial context were presented with different colors on the PDB 3D structures for the structural characterization of PTM substrate sites. Following the method of Ruzza et al., the functional roles and drug binding effects associated with a PTM substrate site’s spatially neighboring residues were determined on the basis of these amino acids’ side chain orientations. As shown in Fig. 2b, given an N-linked glycosylation substrate site and its spatially neighboring amino acid, the vector from the residue to the nitrogen of N-linked glycosylated asparagine (p) is:

$$ {S}_k={X}_p^{SG}-{X}_k^{C\alpha} $$

(1)

The angle θ _k has to be less than 80° for the spatially neighboring amino acid k to be considered a functional residue to the asparagine on the N-linked glycosylation [17, 44].

Case study of PTM sites associated with drug binding

More than 1100 PTM substrate sites related to drug binding were curated and archived in CruxPTM after a large-scale screening for PTM substrate sites and drug-binding sites in the PDB. The number of drug binding associated sites for each PTM type can be found in Table 2. Most of the drug binding sites appeared to be able to undergo phosphorylation, while the second most common PTM among the drug binding sites seemed to be ubiquitylation. According to dbPTM [6], the phosphorylation state of Ser843, situated close to the drug binding site (6.4 Å), could influence the affinity of binding for the agonist and inhibitor of the mineralocorticoid receptor (MCR). This is supported by the observation that posphorylated Ser843 reduces the MCR’s binding affinity for its agonist and leads to the receptor’s own inactivation [45]. Phosphorylation does not always inhibit the protein’s activity. In the case of most kinases, while reducing the affinity between a drug and its target, phosphorylation can actually increase the activity of the target protein [11, 46–48]. The insulin-like growth factor 1 receptor (IGF-1R) is an example of the type of kinases (Fig. 3). A case study of IGF-1R shows that an inhibitor of the receptor could maintain the protein in an inactive conformation; however, if the receptor becomes phosphorylated, the crystal structure of its activation loops would be rearranged in such a way that significantly decreases the inhibitor’s affinity for the receptor while enhancing the activity of the receptor. Therefore, phosphorylation may affect the efficacy of a drug by modulating the structure of the target protein and reducing the affinity between the drug and the target.

PTM Instances	Number of PTM sites located on drug contact sites	Number of PTM sites within 12 Å distance of drug contact sites
Phosphorylation	400	1032
Ubiquitylation	164	331
Acetylation	113	296
Pyridoxal phosphate	84	103
N-linked Glycosylation	82	375
O-linked Glycosylation	51	50
S-nitrosylation	38	47
Methylation	25	92
Disulfide bond	18	34
FAD	17	0
Gamma-carboxyglutamic acid	17	32
Proteolytic Cleavage	17	68
C-linked Glycosylation	15	23
N6-carboxylysine	15	12
Carboxylation	12	30
Nucleotide-binding	8	5
Glutathionylation	6	0
Myristoylation	6	10
N6-succinyllysine	6	21
Palmitoylation	6	26
Dephosphorylation	5	5
Oxidation	5	12
FMN	4	0
Prenylation	4	3
Thioether bond	4	19
TPQ	4	0
Deamidation	3	0
Dehydroxylation	3	0
Glycation	3	0
Isopeptide bond	3	0
Neddylation	3	0
Pyruvate	3	0
Sumoylation	3	47
Tryptophylquinone	3	0
ADP-ribosylation	2	3
Biotin	2	0
Deacetylation	2	2
Hydroxylation	2	13
Lipoprotein	2	0
Nitration	2	10
Phosphopantetheine	2	0
Pyrrolidone carboxylic acid	2	21
Tryptophyl-tyrosyl-methioninium	2	0
Allysine	1	0
Amidation	1	36
Carbamidation	1	0
Chromophore	1	0
Formylation	1	1
Lipoyl	1	0
N6-malonyllysine	1	0
S-linked Glycosylation	1	0
Sulfation	1	3
Transglutamination	1	0
TTQ	1	0
Total	1189	2762

With reference to a case study discussed on dbPTM, an acetylation site (Lys199) on human serum albumin (HSA) is essential for drug transport and metabolism [49]. Annotations on the OMIM database [50] implicates HSA in hyperthyroxinemia (OMIM ID: 615,999) and analbuminemia (OMIM ID: 616,000). Accordingly, this investigation shows that the Lys199 residue is 4.3 Å from the salicylic acid (DrugBank ID: DB00936) binding site. Aspirin (DrugBank ID: DB00945) has been found to be able to acetylate Lys199, while being hydrolyzed into salicylic acid by HAS [6]. Thus, our investigation confirmed the conformational plasticity of HSA and provided a possible explanation for the regulation of HSA-drug interaction through PTM at the Lys199 residue. Figure 4 shows another example of the modulation of drug-target interaction through PTM. Urease is responsible for hydrolyzing urea into carbon dioxide and ammonia. Carbaoxylation of the Lys217 residue on the Klebsiella aerogenes urease coordinates the contact between two nickel ions and the drug molecule acetohydroxamic Acid (DrugBank ID: DB00551). The active site of all known ureases is composed of a bis-μ-hydroxo dimeric nickel center, located in the alpha (α)-subunit, and has an interatomic distance of ~3.5 Å [51]. Our analysis shows that acetohydroxamic acid might inhibit urease activity by competing with nickel atoms in the enzyme to form a chelate. This could potentially interrupt the hydrolysis of urea, which reduces the concentration of urinary ammonia and lowers urine pH.

Functions of PTM sites on protein-protein interactions

As shown in (Additional file 3: Table S3), of all the experimentally verified PTM sites, over 20% can be found in the functional domains of proteins, implicating the biological significance of PTMs. We studied these sites to infer the roles that these PTMs play in PPI interactions. For instance, approximately 70% of known S-nitrosylation sites, which are responsible for the regulation of NO-related cellular processes, are located in functional domains. Also, among the data that we have collected for the current study, more than 1900 PTM sites are localized to the interface of domain-domain interacting regions. Based on our observations, it appears that structural associations exist between many PTM sites and binding sites for specific PPI domains and perhaps even regulate the interactions between proteins by modifying the sites of contact.

Cyclin-dependent kinase inhibitor p21, by binding to cyclin-dependent kinases (CDKs), acts as an important checkpoint in cell cycle arrest in response to DNA damage [52]. It can also bind to proliferating cell nuclear antigen (PCNA) to suppress DNA replication [53]. While in solution, the p21 protein does not exhibit a stable structure. However, upon binding to target proteins, the protein assumes an ordered stable conformation. Figure 5 shows that phosphorylation of the Thr145 residue of p21, which corresponds to the PCNA binding region (from residue 144 to 151), may inhibit the interaction between p21 and PCNA, resulting in PCNA binding with other DNA polymerase components [54].

The Rho GDP-dissociation inhibitor 1 (RhoGDI1) is a regulator of the Rho family GTPase [55]. By preventing the release of GDP and the loading of GTP on Rho proteins, RhoGDIs can inhibit the activity of Rho family GTPase. RhoGDIs can also block the degradation or improper activation of inactive Rho proteins by transferring them to the cell membrane. The function of RhoGDIs is regulated by phosphorylation at their Ser, Thr and Tyr residues. In fact, phosphorylation at multiple sites in RhoGDIs can stimulate the simultaneous release of multiple Rho proteins [55]. The key functional region of RhoGDIs lies in its N-terminal domain. This region, though generally disordered, can form two helices and bind to the switch I and switch II regions of GTPase to prohibit the latter from making conformational changes required for the exchange of GDP and GTP [56]. As presented in Fig. 6, the disordered N-terminal domain of RhoGDI1 contains a Tyr residue (Tyr27), which is localized to the binding interface and can be phosphorylated to facilitate the dissociation of RhoA, Rac1, and cdc from RhoGDI1, making GTPases available for activation [57].

The crystal structure of the ternary complex of the eIF4E-m7GpppA-4EBP1 peptide is shown in Fig. 7. This structure, formed from the interaction among the 7-methylguanosine at the 5′-cap of mature transcripts, eukaryotic initiation factor 4E (eIF4E), and endogenous 4E–binding protein 1 (4EBP1), is required for translation initiation [58]. Figure 7 indicates that three substrate sites (Thr50, Tyr54 and Ser65) of phosphorylation are located within the binding region of the 4EBP1, which can regulate its interaction with eIF4E. The three sites are reported to modulate the reversible binding of 4EBP1 with eIF4E, and hyper-phosphorylation at these sites can decrease the strength of interaction between the two proteins [59].